Ensure gmic display works in Jupyter

[myselfhimself]

Nice to have - result image display

[myselfhimself]

Possibly:

1. detect non-interactivity / IPython environment
2. use unconfined IPython display as in this StackOverflow answer using either a temporary file address or a base-64-encoded image.

2. tested successfully on a Binder Jupyter.

[myselfhimself]

Jupyter detection techniques:
https://gist.github.com/DIYer22/5cc015dcef53d62c16bdff0f8f345e96

In short, detecting a gui (be it web or Qt) boils down to: 'ipykernel' in sys.modules == True. Tested in jupyter qtconsole (True), jupyter online (True), Google Colab (True) and ipython (False)

[myselfhimself]

We could, depending on availability on the client's Jupyter notepad:

1. use the matplotlib interactive viewer if available... without using the to_numpy_array() converter for now, but just an image output (b64 or temp file)
2. use IPython's image display

[myselfhimself]

"DISPLAY" key existence-testing in sys.environ helps to know if gmic.run("display") will fail display anything for sure (tested) https://stackoverflow.com/a/8258144/420684

[myselfhimself]

The pure-Python implementation could be like:

def gmic_run(command, images):
    # image processing .. changing 'images' in place
    if "DISPLAY" not in sys.environ:
        for pos, image in enumerate(images):
            #get image base64 or have it saved to temp files in the same gmic.run() run or another one
            try:
                import matplotlib.pyplot
                if image_names and pos < len(image_names):
                    pyplot.title(image_names[pos]) #use gmic image names if available
                #save image or make it b64
                pyplot.show(image_saved_or_b64)
            except:
                try:
                    from IPython.core.display import Image, display
                    display(Image('image_path...', unconfined=True))

Will implement something not today

[myselfhimself]

the "display" command within a jupyter / ipython / X-less environment will (now just with a nodisplay=True flag in gmic.run.. because my computer has display) dump images to local temporary directory under unique id png filenames and display with:

1. matplotlib - if available
2. ipython.core.display() - if available

This implementation with temporary files instead of base64 has yet to be bulletproved for web interfaces... I shall try with jupyter in a local web browser.

calling display multiple times in a same command works OK and all generated images are used for display
G'MIC provides display without parameters (supported), with a prepended dash -display (unsupported) but also with parameters (unsupported, would error), display3d (unsupported for ipython/jupyter and will result in a parsing error for now)..

My goal here is not to do a full fledge support of all ways to reinterpret the display command... I might introduce a flag for auto-display as well, so that any final image gets displayed without having to use a display command as well...

[myselfhimself]

Coming soon: matplotlib support - inline or with the qt5 widget..

[myselfhimself]

Matplotlib support is now there with multiple images working:

[myselfhimself]

• test in python console
• test in ipython console
  • prevent wurlitzer macro non-enabling message on import gmic
• test in jupyter qtconsole
• test in jupyter notebook (local)
• test in jupyter notebook (online)
• test in google colab (online)

[myselfhimself]

Now works well without any command to write for UNIX users. See #64 (comment)

[myselfhimself]

In Google Colab, after the 2.9.1-alpha1 release:

[myselfhimself]

In Jupyter Lab online:

[myselfhimself]

A pure ipython console, in a non-web setup, with DISPLAY environment variable unset, does not need to use wurlitzer to show console output, still there is a warning on G'MIC import, which could be omitted or clarified if we detected a pure IPython console.

The G'MIC display is used instead of any matplotlib with DISPLAY set, and there is not display at all, if DISPLAY is unset.

Adding TODO checkbox above for fine-tuning this import-time message regarding wurlitzer in IPython console.

[myselfhimself]

Wurlitzer's code shows that when IPython has no kernel available, we are in within a terminal Python context, and wurlitzer itself is useless.

[myselfhimself]

Good enough, will not be automatically tested, in order to give more time for other tasks

[myselfhimself]

• fix non-displaying gmic.run("display",gmic_image_or_gmic_image_list) on Google Colab
  Related: https://colab.research.google.com/drive/1GNaAql2ZMwJ-8TL54jnpgnoHciCuIqtW?usp=sharing